Document nk9vBkxqKo19wvYqZZmGO5O2w

I UNITED STATES ENVIRONMENTAL PROTECTION AGENCY WASHINGTON, D.C. 20460 March 4, 1983 Dr. David Weil Environmental Criteria Assessment Office US EPA Research Triangle Park* N. C, 27711 Dear David, I am enclosing the memo I have written that I would like the reveiw panel to read before the meeting on next Thursday. 1 have also enclosed a review by Dr. Dan Violette and Dr. Craig Miller which I would also like shared with the panel. The copy I am sending you is their first draft. They will send you their final draft separately on Friday, I am enclosing this copy as a backup in case the second one is inexplicably delayed. Thank you very much, I will see you on Thursday. Sincere Joel Schwartz, Ph.D N33861 TEH 0532917 UNITED STATES ENVIRONMENTAL PROTECTION AGENCY WASHINGTON, D.C, 20460 MEMORANDUM SUBJECT; The Use of NHANES II to Investigate the Relationship between Gasoline Lead and Blood Lead FROM; Joel Schwartz Office of Policy Analysis TO: David Weil Environmental Criteria Assessment Office You have asked for any additional comments on whether the NHANES IX data set can be used to examine the relationship between gasoline lead and blood lead. This memo will focus on that issue, and I will subsequently provide a more detailed memo amplifying on our study. Since the issue of whether you can analyze the NHANES data and how you analyze it are inextricably mixed, and since it / involves not merely theoretical issues but pragmatic ones including the strength of association, I will also discuss our analysis here. Our basic contention is that while the NHANES II data are not optimal for studying this issue, nor most efficient, even an in efficient 10,00 person sample leaves the possibility of analyzing the relationship, if properly controlled, and if the strength of associa tion is large enough. We believe it is possible to control for the imperfections in the sample for these purposes, and that the associa tion is demonstrably strong enough to resolve the issue. We also believe that the question of how rigorous a test these analyses are put to must be viewed in context. This context requires examining N33861.01 TEH 0532918 DUP05003418& ---- -2 the rigor to which the air lead/blood lead studies are subjected# and to which some of the medical effects studies are subjected# so that there is some consistency in the degree of statistical igor required by the criteria document# and so the panel examining the NHANES study understands the degree of uncertainty the Agency has implicity indicated it is willing to accept. Issues of Sample Control The principle Objection raised to using the NHANES II data is that it was drawn sequentially in time# by two stands# each spending two to three months at a site# and moving to another location. Thus, the representation of different subgroups changed Over time as the two sampling locations changed. While this is clearly a problem# the extent of the problem is a function of how the regression analysis controlled for this# and of the pragmatic issue of whether those subpopulation had different lead levels, or different responses to gasoline lead. By including dummy variables to account for level effect changes attributable to different demographic subgroups one can \ prevent changes in the mix of those groups from causing changes in blood lead which the regression might otherwise attribute to gaso- In our multiple regression analysis we controlled for three age groups# three income groups# rural areas, cities, and large cities, whether city residents lived inside the center city or in suburbs, sex and race. Industry critics of the first analyses CDC did of the NHANES data had emphasized that since children and city residents have higher blood lead# it was important to control TEH 0532919 DUP050034189 -3- for changes in their representation. The above dummy variables appear to control for all the major demographic factors that have been associated with differences in blood lead levels. Incomer race and outside center city (which correlates well with post 1950 housing), should provide Indirect control over lead paint exposure. Food lead, as discussed in our paper, increased over the period, and therefore the gasoline variable will not pick up any drop in blood lead that should be attributed to decreased food lead. Interaction terms also play an important role in controlling for subpopulation changes. In addition to testing for and including, significant interactions between demographic dummies, we tested for the interaction between main effects and gasoline. This allows us to control for changes in the representation of groups that may have different responses to gasoline lead, and avoid biasing the gasoline coefficient. It is worth noting not merely that we controlled for this, but that the pragmatic results were telling. /Jrf ^Rural gas was statistically insignificant, i.e. inhabitants of --rural areas had the same change in blood lead as gas lead changed as urban residents did (Of course absolute rural levels were lower). Inside center city was likewise not significant, and for A whites, children had the same rate of change as adults. This ^ increase our confidence that changes in the sampling of NHANES i over time do not invalidate our results. Finally it is worth mentioning that during the survey the two stands were always kept in different regions of the country, that in general each stand X; tested urban and rural residents of the county it was located in, and that they moved frequently enough, relative to the rate TEH 0532920 DUP050034190 r* i - 4 --p of change of gasoline lead usage# that the sample was not as unrepresentative as some have implied# even before the use of dummy variables. Another important control we examined was to include time and season as explanatory variables# to see if they could make gasoline lead insignificant. It is important to note first that there is c>>no outside evidence that any other major source of lead was diminish"* d ing during this period. Food lead increased and lead paint exposure changed little. Thus there is no theoretical justification for including time# it: was merely examined to validate or invalidate the association with gasoline. Including time puts gasoline's ex planatory power to a strenuous test# since it must compete against another variable that is highly correlated with itself and blood lead. Gasoline can do this by differing from time in its fine structure. (Gasoline lead did not change ^consAantlv. among other O things it has a strong seasonal component.) To further test its explanatory power# however# we included seasonal dummies which could compete with gasoline's short term behavior in explaining blood lead. In addition# we simultaneously examined the lag structure for gasoline by including variables for two different months gasoline lead usuage. Despite this# lagged gasoline lead was significant at better than the 97% confidence level (^=^34^). Moreover since we were testing the two gasoline Variables jointly we should really compute a joint F test for them that would be even more significant. hen season (which was insignificant) was dropped# the F value for lagged gasoline rose to ^^22. Thus even controlling for hypothetical changes over time and season as well as the main effects and inter actions# gasoline was significant. These results are included in TEH 0532921 DUP050034191 -5- Appendix I including both the weigted OLS results and the Surregr results which adjust the variances for the stratified sampling by using an algorithm based on 32 pairs of psuedosampling units. Finally, and perhaps most telling, we have inserted dummy vari ables for the sampling locations into the model to account for any variations in blood lead levels due to the specific sites themselves and not controlled for by the demographic dummies* the results of this regression leave the gasoline lead variable with a t statistic of 19 (before Surregr Correction), and the coefficieint of lagged gasoline lead is essentially identical to the sum of the coeffici ents for current month gasoline (PBNAT) and one month lagged gaso line (GASLAG) in the earlier models* It also attributes\57%\of the mean blood lead over the four year period to the mean gasoline lead usage. We used only one gasoline variable in this run because of the large size of the model. We also only used 48 variables for the 48 site locations NCHS has released. The remaining 16 locations have a population less than 100,000 and their location has not been released for reasons of confidentiality. We have represented them all by one additional dummy. This result is also included in Appendix I and indicates that neither variation in subpopulation representation nor the choice or pattern of sites themselves interferes with using the NHANES II data to determine a relationship between gasoline lead and blood lead. Empirical Observations The NHANES data cannot be examined in a vacuum, we must use our knowledge of the etiology of lead exposure, and theoretical TEH 0532922 DUP050034192 models, along with statistics. This other knowledge bears directly on the ability to use the NHANES data. The absolute magnitude of the drop in average blood lead levels, 5-6 ug/dl, severely limits the number of variables that can be causally related to the drop for the simple reason that few factors contribute as much as 1 ug/dl to blood lead. The possibility of mispecification error is inherently limited, therefore. The other principle factors that contribute large amounts of blood lead, and which are capable of changing, are food and paint.1 As mentioned above# food lead intake did not decrease over the period. Paint lead is an unlikely explanatory variable for several reasons. First, paint is a more significant contributor to high blood levels than to low ones. This is because exposure to paint lead usually results in the absorption of enough lead to produce large increases in blood lead. However, the NHANES II drop in mean blood lead involved the entire distribution shifting. It was not due to effects primarily concentrated in the upper end of the distribution. Indeed, even low blood lead groups had significant declines in blood lead. Second, adult blood leads fell as well as children's blood lead, and adults generally do not eat paint. Third, the drop occurred in suburbs of central cities. These have a lower fraction of pre-1950 housing stock, and therefore inherently less exposure to lead paint, yet they showed the same drop and the same gas lead coefficient. Finally, lead paint removal ISoil lead due to years of buildup may contribute substantially, however the only change in this factor Would be the change in most recent dustfall. This dustfall is principally due to gasoline, which contributes about 90% of emissions of lead to the air. TEH 0532923 DUP050034193 programs during this period reached only 50,000 of the 30 million housing units with lead paint. Since there do not seem to be any omitted variables that could have accounted for any significant fraction of the large magnitude change we saw in NHANES, we can be more confident about using the data for this type of analysis. Another empirical fact supporting the use of the NHANES data is the large seasonal variations present in both gasoline lead and blood lead. Dr. Billick's data (from the New York, Chicago, and Louiville screening programs) also verifies that this seasonality has existed in blood lead for years be fore NHANES period. This provides an opportunity to test gasoline's explanatory power over two different scales, seasonal fluctuations and long term trends. Moreover, the causes of the increase in gasoline lead in the summer is a volume effect due to an increase in driving. The cause of the long term drop in gasoline lead was a regulatory program that effected density [grams per gallon). The fact that we can test the relationship between two different time trends caused by different factors makes the NHANES data more reliable than might otherwise be thought. And the fact that season's are statistically insignificant when gasoline is introduced, means that there is no statistically significant difference between the short term and long term blood lead to gasoline lead coefficients. The fact that these slopes are the same is an important verification of the procedure. It also speaks to mispecification. It is difficult to believe, for instance, that the amount of lead paint in housing oscillates seasonally. Moreover, in the summer children spend less to indoors, exposed to interior paint. While there is some lead paint on the exterior of housing, it is indoor lead paint that has TEH 0532924 DUP050034194 8- been linked to undue lead exposure* In addition, much public housing has brick exteriors. Additionally, the lag structure for gasoline provides some empirical guidance on the ability to use the NHANES II data. If gasoline lead is merely a proxy for time, or some artifact of the sampling path we would not expect an examinst-ion- of its--lag structure to yield something meaningful. In particular, current gas lead, one month lagged gas lead, two month lagged gas lead, etc., should all be equal proxies for missing time. In fact, the lag structure agrees well with what we would expect physically. Current and one month lagged gasoline lead are ^cyyjyc|mr, with the one month lagged gasoline having the most significance and the highest coeffi cient. The latter lags fall off in significance and coefficient size. This agrees with the view that gasoline lead in dustfall through resuspension in the air and injection, is the major pathway between gasoline lead and air lead. This lag structure is not consistant with the view that gasoline lead is simply picking up a time trend caused by some unexplained change in an unknown other factor, nd. pendent of the fact that no one has been able to suggest a reasonable other possible cause for the fall in blood lead* It is so difficult to believe that the time pattern of the surveying could have produced this structure. Finally, there are the confirmatory studies that put the use of NHANES in perspective. First, it should be noted that the lead isotope study in Italy removes any doubt that gasoline lead contri butes large amounts of lead to blood. By changing the isotopic composition of the lead added to gasoline in the Turin area and TEH 0532925 DUP050034195 9- looking at isotopic changes in blood lead/ Pachetti and others discovered that gasoline contributed 6-8 ug/dl to adult blood lead. Moreover# blood lead in adults constantly re-equilibrates with large stores of lead in bones. Studies by Rabinowitz indicate that it takes more than tour months^{(for blood lead isotopic ratios to correctly reflect current exposure rather than historic isotope ratios stored in bone. Therefore Pachetti et el concluded that they may be underestimating the contribution from gasoline. Children equilibrate much faster since they have less bone lead stores. Unfortunately the Turin study had no baseline ratios on children. If one assumed children had the same baseline isotope ratio as adults this would suggest substantially higher contributions from gasoline. These results are in very good agreement with NHANES regression# which also predicted 8 ug/dl from gasoline. Of course Italy is not the U.S. (on the one hand in 1976-80 Italians used no unleaded gasoline, whereas about 34% of U.S. gas was unleaded then. On the other hand U,S. cars got only about half the mileage of Italian cars.) Nevertheless the conclusion that gasoline contributes significant amounts to blood lead# on the order of 5-10 ug seems reasonable to apply to the U.S as well as Italy, In addition to evidence confirming the reasonableness and approximate magnitude of the general results, there is evidence that helps lay to rest concern over the way the NHANES survey moved around the country. For example, Rabinowitz and Needleman analyzed lead levels in the blood contained in the umbilical cords of 11,837 birth between April 1979 and April 1981 in Boston TEH 0532926 DUP050034196 * 10 lying in hospital. When they correlated these with local gasoline lead emissions, and examined the lag structure, they found a significant correlation (P*0.001), with the one month lagged gasoline lead having the strongest relationship, as analysis of the NHANES data also indicated* They found 57% of the variance in blood lead explained by gasoline. There were no significant trends of smoking, education level, or alcohol consumption with blood lead levels. Water lead levels appeared to rise over the period. The ICF report on the NHANES data also shows the results of regressions on the blood lead screening data in New York, Chicago, and Louisville, done separately for blacks and whites, V Again gasoline lead is a significant explanatory variable explaining most of the drop in blood lead. Again season is insignificant in these cities, indicating that gasoline explains seasonal as well as long term trends. While these children are not a random sample of the population, approximately 20% of all black preschool children were screened each year, making it unlikely that variations in sampling produced these results. These three city studies confirm that even if you stay in one location, the strong relationship with gasoline holds. Thus the regression results are unlikely to be significantly affected by the way the sample was taken. Similarly CDC tabulates summary statistics for those screening reported to them each quarter. There are major changes in the level of screening in each city over time, as well as in which ones reported, so this aggregated data is not very useful for regression purposes. However, there were two discontinuous changes in EPA's lead rules TEH 0532927 DUP050034197 -11- in late 1979 and late 1980. The enclosed graph shows the percent of children with lead toxicity nationwide, evidencing a clear sign that the two regulatory changes produced drops in the percent of toxic children. Therefore what happened in N.Y., Chicago and If Louisvillejappeared to occur nationwide, and discontinuous changes^ in lead rules- had perceptable impact. The accompanying graph for Providence shows the results are not a function of variations in cities reporting. Again a study by the Illinois Institute for Environmental Quality of the city of Morton Grove, an upper middle class suburb (population 28,000) west of Chicago, consisting primarily of post 1950 housing found that children living within 200 feet of a major roadway had a statistically significant (pCO.Ol) higher mean blood lead than children living more than 200 feet from a major roadway. For children living within those boundaries all their lives, the difference in mean blood leads was about 5.6 ug/dl. It is also interesting to note that in 1974-75, when gas lead emissions were very high, most of these white upper income children (living in homes expected to be free of lead paint) had blood lead levels over 20 ug/dl range. By the NHANES period similar children had signifi ed. ..My lower lead levels, and gasoline lead usuage was also down significantly. Consistency As mentioned before, it is important for the criteria document to apply consistent statistical standards across the body of data it examines. We recognize that uncertainty is inherent in evaluating TEH 0532928 DUP050034198 TEH 0532929 DUP050034199 *X 3 J H E I t Ho a -j 31 TJL 7 t !>' % nj o : TEH 0532930 DUP050034200 - 12 - data on the effects of a pollutant/ and judgements must be made on what is our best estimate based on available data. Since industry representatives on the workgroup have suggested subjecting the NHANES analyses to extremely rigorous examination/ it is fruitful to examine how other data already included in the criteria docume have met these tests. Medical Data Medical data is clearly of a different nature that epidemiological data, since it examines detectable biological changes* Nevertheless, in order to get an ambient standard EPA must frequently know how much of an effect occurs, and at exactly what lead levels. Questions of population representativeness and sample size are clearly relevant here since different groups respond differently, and because even homogeneous populations have distibutions of responses to environmental stress, and small samples may misrepresent the size of the response. The health effects section of the criteria document nevertheless relies on studies that frequently involve small and non representative samples. This is for the excellent reason that those are frequently the only studies we have, and the agency cannot ignore the results. Nevertheless this procedure cannot be ignored in evaluating the use of dose response studies using the NHANES II data. One example of this approach is a section of chapter twelve (P24 and 25) discussing the affects of lead on inhibition of an enzyme (PY5N), Papers are cited that studied 9 adult male workers, and 21 children, among others in drawing conclusions about the degree of inhibition. Obviously larger samples more representative of the H.S-. population TEH 0532931 DUP050034201 .4 > ...i***^*-*-- -13- would have been desirable but clearly conclusions are capable of being drawn from the existing studies* Air lead/Blood lead studies Clearly these studies are more relevant since they deal not with medical effects but blood level response to environmental expo sure as the NHANES regressions did* They include {Chapter 11) Griffen's study of 14 adult male prisoners exposed to lead aerosols in chambers, Rabinowitz's study of 5 adult males volunteers simi larly exposed, and Kehoe's study of 12 subjects living in the vicinity of a lead smelter. Prom these and other similar studies of small nonrepresentative populations, EPA derived air lead to blood lead ratio's which were used in 1978 to set the ambient standard* They also receive prominence in the draft revised criteria document. A graphic representation of their results is included (Figure 13-2). The real question that the panel should consider is whether regres sions between gasoline lead (90% of air lead emissions) and blood lead, using the NHANES data, has anything to contribute to the knowledge of the relationship between air emissions and blood lead levels, and whether the three studies cited above have been put to the same stringency of examination. TEH 0532932 DUP050034202 <***.*m-,.*_*. PRELIMINARY DRAFT figure 13-2. Blood lead vs. air lead slope as a function of maximum blood lead f 313B/A 13-28 1/03/83 TEH 0532933 DUP050034203 -14- Appendix 1 Multiple Regression Analyses of NHANES II Data N.B. These show the weighted OLS results performed by Surregr, a software package designed to properly analyze stratified samples, while the coeffici ents are correct/ these results overstate the F statistics because of oversampling of extreme groups. To correct thisf Surregr next constructs 32 pairs of psuedosampling units and estimates the correct variances. An example of one of these second stage runs is included. All the results in our paper use these final runs, the first stage is displayed here to resolve ques tions that were raised abut whether we analyzed individual or grouped data. TEH 0532934 DUP050034204 lr.iTvirr.s2v ic: u*v.' ntswvu* v lZ`W *ru..iNrtTir. IEwt.e; $r Ffittva* arrmficN; t'M>.) r VrfcAs i'zvr^Lv l t rrmvit rhvptriun lAIi TESI FGn OVERALL ROISl OlfOErtrilEKT SrrtvTS TIIE tern! TS-nttii: KirKIICAS fiOLIHALF KALI RURAL RURwJGAS SHALL IKCCtEl IHKfrE2 KIK7EF, SfRINO f al l 17*. vaa* 26.16 11.6$ 7,73 3.5E 0.14 73.64 1.87 3.67 0.05 25.05 42.46 21.25 C.S3 5. CD 1,06 17 0.0505 1 v. y\w* /.a#.a 1 v.C\vc- 1 0.0054 1 0.05E:1 C<"AA. 1 0.0500 2 0.171$ 1 0.0556 1 1 C.P2S0 CiWa aWKa 1 0.0000 2 0.0000 1 0.341? 1 . 0.0556 I 0.302? o #v OASLA3 5h k \.` h - uiU ^s Fr*AT /*Ai lo.OS 6.64 (aCa\VAwr. TEH 0532935 DUP050034205 IKtrOTHESlS 7ES7IHS RESULTS o es s wmimm deg r ees d f f r eedst , CSPSeftl EFFECT F VALUE KSfiEFf fif FREEDOM PROBABILITY LEAD TEE* FOR G.mL MODEL 0IS5EF333C EFFECTS 71TZ TEE!TFE5&&E !::iKIS6SS AULTRALE KALE RURAL RUKALGAS SMALL IfCOMEl INCOMES GA3LAG FBNA7 25S.7B 33.S2 11.43 7.75 S.S5 C.1S 73.83 i.s: 3.47 C.OS , 30.02 4.1,51 21.28 50.17 13.15 > 14 Lt:x 1 t.vvD 1 C.'v% 1 1 1 L.i':: 1 . 1 C.ITtS 1 v.vcli 1 0.7SC? 1 C.5000 1 v.0000 1 C.GOOO 1 0.0000 1 0.5552 iHYFCTnES;? TESTINB FESULTS C o5S? BSOTKATGR rKFBB&ff EFFECT DEGREES .OF F VALUE FREEDOM DEGREES o f FREELY PROBABILITY LEAD TEST FOR OVERALL K5KL 245.75 OJAI-'E^aiT EFFECTS TIrE TEEN TEEN"!ALE k :i 31. SB 11.4c /V %' , * .* ' ' 12 0.0500 1 0.055/ 1 G.C0C7 3 c.r:: 1 V.Wvv AT.TMLE --E F.5KA. BKALL InDuEl IfC0E2 6KLA3 fr.:;-.T 71.23 i. JO- 75. AT SO. IS 43.57 4**..i -. 50.25 35.27 3 5.0050 1 0.37c- 3 5.5505 3 0.0555 3 0.0500 3 5.0500 3 r.ocvo 3 v. v 553 TEH 0532936 DUP050034206 jtwcTMrsii "T.TS - SToi pti:>w:t:v. t-KF?ES7 r?FEn IcOStr r (r f1tEiWJ F VoliS lEBREr * O' rr,:FIv:> FROSAEIITTY LEAL TEST Fur; OVERAIL KOSEt OIUErEKDEW EFFECT? TEcK TEEMK1 KIP KTD5A2. A5L7KALE KALE &RS. WMLG/tS SHALL JUC&E1 lfcu*E2 GASLAD FA7 222.70 31.16 7.52 3.70 0.14 73.es 1.83 3.26 0.32 41.31 48.30 25.30 47.74 37.64 0,0000 0.0008 0.CM9 0.0545 I C.70c5 1 0.0000 1 0.1765 1 0.0712 1 0.5471 1 0.0000 1 0.0000 i 0.0000 i 0.0000 i 0.0000 TEH 0532937 DUP050034207 CARY. N.C. 27511-8000 1 0 0H0BEL: nmol SSE DEE DEP VAR: LEAD HSE s t a t is t ic a l 209209.9 8542 24.491916 F RATIO PROBF R-SQUARE ANALYSIS SYSTEM 54, 66 0.0001 0.2673 20U6 THURSDAY, MARCH 3, 1583 0 VARIABLE IHTEKEFT TEEN KID ADLTMALE MALE RURAL SHALL INCONEl INC0NE2 P8NAT1 CITY2 CITY3 CITY4 CITY5 CITV6 cir.7 CITY8 cm? crmo CITY11 CITYI2 ClTi'13 CITY14 CITY15 CITY16 cmi7 CITY18 cmi? cm'2o CITY21 CITY22 CITY23 CITY24 CITY25 C1TY26 C1TY27 CITY28 CITY2? C1TY30 C1TY31 CITY32 CJTY33 C1TY34 C1TY35 C1TY36 CITY37 CITY38 CITY39 PARAMETER STANDARD IIF ESTIMATE ERROR 1 2.414253 1 0.097845 1 2.948726 1 3.656996 1 1.121172 1 -1.083656 1 -0.606628 1 1.887866 1 0.889920 1 2.075042 1 -0.953462 1 0.073215 1 -0.523587 1 2.403546 1 2.706952 1 0.939CS5 1 -0.593643 1 -0.163830 1 -0.041216 1 -0.891395 1 1.574945 1 -0.143501 1 4.543707 1 0.887415 1 0.911101 1 -0.570510 1 0,964078 1 2.841471 1 -1.921417 1 1.288683 1 -1.767744 I -0.774917 1 1.562597 1 0.377149 1 -1.133393 1 3.382883 1 2.637064 1 0.052393 1 2.420480 1 0.930684 1 1.007609 1 0.399083 1 -0,247098 1 0.277434 1 0.422788 1 1.558622 1 1,678821 1 0.855217 0.595597 0.185788 0,174653 0.219709 0.170369 0.386807 0.396211 0.163280 0.119018 0.104640 0.460041 0.695473 0.644479 0.589658 0.665278 0.472427 0.481660 0.524241 0.556119 0.429897 0.342878 0.439475 0.466142 0.606444 0.508252 0.430307 0.637285 0.555503 0.433118 0.623215 0.54017? 0.581860 0.398322 0.660972 0.674583 0.631244 0.72784? 0.678400 0.503935 0.614336 0.600625 0.617828 0.420854 0.425473 0.469418 0.563003 0.681264 0.574295 T RATIO pr o b>;t j 4.0535 0.5267 16,8834 16.6447 6.5808 -2.8015 -1.5311 11.5622 7.4772 19.8303 -2.0726 0,1053 -0.8124 4.0762 4.068? 1.9877 -1,2325 -0.3125 -0.0741 -2.0735 4,5933 *0.3265 9.7582 1.4633 1.7926 -1.3258 1.5128 5.1151 -4,4362 2.0678 -3.2725 -1.3318 3.9230 0.5706 -1.6801 5,3591 3.6231 0.0772 4.8032 1.5149 1.6776 0,6459 -0.5B71 0.6521 0.9007 2.7684 2.4643 1.4892 0.0001 0.5734 0.0001 0.0001 0.0001 0,0051 0.1253 0.0001 0.0001 0.0001 0.0382 0.9162 0.4166 0.0001 0.0001 0.046? 0.2170 0.7547 0.940? 0.0302 0.0001 0.7440 0.0001 0.1434 0.0731 0.134? 0.1304 0.0001 0.0001 0.0387 0.0011 0.1630 0.0001 0.5683 0.0?30 0.0001 0.0003 0.9334 0.0001 0.1298 0.0935 0.5183. 0.5571 0.5144 0.3678 0.0056 0.0137 0.1365 VARIABLE l abel ** iHittri* 7*.* ^ 4 v {**.< . X / TEH 0532938 DUP050034208 1 3 ) VARIABLE CHY40 CITY41 C1TY42 CITV43 CITY44 CITV45 CITY46 PARASTSI IF ESTIMATE 1 2.967630 1 0.787219 i 0.589933 1 2.555038 l -0.450362 l 0.383373 J 1.255387 STATISTICAL ANALYSIS SYSTEM STANDARD ERROR 0,489727 0.516268 1.048507 0.455331 0.382247 0.452761 0.497420 T RATIO 6.0598 1.5248 0.5626 5.6108 -1.1782 0.8467 2.5238 FROBMTS 0.0001 0.1273 0.5737 0.0001 0.2388 0.3972 0.0116 . VARIABLE LABEL / 20:46 THURSDAY, MARCH 5, 1923 CITY47 CITY4S CITY49 1 0 1 1.751771 1 -0.794825 1 1.719334 0.387244 4.5237 0.382457 -2.0782 0.437736 3.9278 STATISTICAL 0.0001 0.0377 0.0001 ANALYSIS SYSTEM 20:46 THURSDAY, MARCH 3, 1983 TEH 0532939 DUP050034209 S u r r *~fr resw/tfs Model Number A Dependent Variable: Lead 32 Denominator Degrees of Freedom Independent Effect Coefficient F Value Degrees of Freedom INTERCEPT TEEN TEENMALE KID ADLTMALE HALE RURAL SMALL INCOME1 INCOME2 NGASPB NGASPB(-1) 3.74 -.63 1.54 3.11 4.35 .66 -1.45 -1.00 1.21 ,59 .74 1.25 8.04 19.20 93.94 169.23 4.74 14.91 7.74 16,50 16.14 4.54 9.17 1 1 1 1 1 1 1 1 1 1 1 TEST FOR OVERAL MODEL 168,26 11 Probability 0.0079 0,0001 0.0000 0.0000 0.0369 0.0005 0.0090 0.0003 0.0003 0.0410 0.0048 0.0000 ICF Incorporated TEH 0532940 DUP050034210 TABLE 2 LEAD IN THE DIET Fiscal Estimated Lead in the Diet in Micrograms/Day Year Infants Toddlers Males Aged 15-20 1973 1974 1975' 21 26 197$ 21 30 1977 `22 28 1978 25 35 1979 36 46 19B0 60.4 90,2 67.2 71.1 79.3 95.1 81.7 82.9 Source: Bureau of Foods, FDA, "Compliance Program Report of Findings, Total Diet Studies." FY 73-FY 80. Note: Blanks indicate data not available. TEH 0532941 DUP050034211 -15Appendix II Umbilical Cord Analysis? Correlation with Gasoline TEH 0532942 DUP050034212 i m: s 1 \K\.'AK'i j '>> !5 jbv.as-c. PJTROl. li'Ml SAMIS AM) IMlMMt I.AI. COIll) IILOOJ) IfcAD UiVELft IMIUMON, MASSAOIl'SXvTni I'rwti.sMy a" tin aittvm iced in A"t:wa otic* today femes ftem lead additives in petrel.1 At: Warn* IfaJ has. Ken traced to bunta:, Hood.1 and rased irvta. of j.rh'rnc lead itau Ken atsixtit'.d with increased blood le.d lryel> in nunv populations of art an dwcllet $, workers. and children.' Chat yes in blnei .cad over time in children rorrrlaie well with nutr.ihly rh-mpt-v tr. the consumption cl leaded petrol..* Wr n per: here on the positive cnrrrla-wvi l-tivui! 3rn !> ot lead ill tiiiibi.i.eid Curd blood and I he oar i'l' leaded jvttnl lit Mjsea.'iiuni*. I,cad levels wric ittriiMind bn 1IHS7 Kith* K.oee.i .April. 197V and Apul. lifel at the Ho mo . I.ymglit Hos-piud (n<nv Brigham and Votnen's Hospital), lllood lead levels displayed both a definite downward ttend w::h mnv and a seasonal pattern.1 Local motorcar use (and petrol consumption) increase) in the sum..:*,- and use ofh-adsddifr.rs has beenchanging recently in resj'oiii- le> Federal standards. We iutve us.J pairwise pudueiiu.-ni.ti-, .;i cut urns of the rnnitlhK cord lead (imans ol aKtut *>1X1 births each} and the tnutv.hly amounis eifpetrol lead sold as reported by commercial sources. tihtnd level (pr d I* 2 Vh* 3 !h\. wt-tii >. :s tl;.- ativ. t.t rf tr.d ot pasoiine (in in' e y*-. -t-iti.j s u m! u. M issa.hjsi. * Vi ot " te sariar.ee ir. nteai , o vj h.-a .In.Is a;csr.nd tc: by tni> cr,e sariallie Th, may cvcrvsttmaic to soinv degree tin so-pc e! the tikic asses.i.u<>n, because rxpnsutc to outdiMr an in lie- tun is higlict in the suttitticr tiotn i:<tet>pen windems anj lm*.c S|H-ni ,vj:J.k '.'s Su . attnbuling S7"i u,~tin vatuncr solely to casohric ieaJ sales w.uid not take into account the confeundinp stcwtul changes tn I'tnxtiiutv nt the pfgiunt wui.n-n !> aii'i'T.i.l de exluiust We round no s.ei-.ifiiat.i (pad-)) ntunthly limtuatfans or tteri in n-an value v those iiutentai fii.iuts su.h as n>lweit> Sfuduttg. edit- Jittin. nt ul.oK! .utstuitpiton win. h wc catjier ifi-inonstrctcd to Is t'tuh.tier i>i !-.) lead Irtel* in this {uiiiiiUirt.n Jap water iv-llcncd ' the hotuvs of 34V of the children at 1,6, .and 1ft months showed a pt.dual fisc in lead itvela nfaKiut 25'v per year, beginning at 5 tap 'I in WSd. a low level. These observations on the close concordance between blond lead and MMtljne lead prompt us tn recommend further reductions ip jllnw.'hlc patolinr Irad s'nn>rni tobnnp blomt trad conccnirction* rvrr. closer to those prevailing tn the pic-industrial era. Il*rvrj Jktisipf. CkttdicnsHMrtiAl Mr&.a\ {>f)'rf NU**vhu*nr*a:i |4.f S A MtOlAtJ. Ha H|NO IT7 PtpttuiKMft wi nJ.r4MiK** PrAtttttnUnretUh.orftn!'mmiyhhii*|h**. JIeRHE*T L. NKEDI.EMAS M onrlds mean umbilical card blood Itad level (Miami total lead matt |r*dtt of gasoline aold in MuwcSuittlt ( ). Tht figuie displays the monthly value* and the table shows the correlation coefficient!, calculated with a variety of lap-times between pans of monthly gasoline lead sales and bloodlcad levels. Significant correlations were observed when pasoiine lead sales in one month were matched with blood lead data Sir the same month, fo: the iivv: month, and for the month after that. Although both variables etc generally declining with time, uie* of '--13 months pave much iWHUct ( Ottclalittns The i>c:.t titling linear lOtn-fatinn is t i'.utiu'Til t u*a-ti,ru<rait-Ciutara:lfaiRoomrv.* >ilioo .(- *I *>- -* K'l IVrs.lt: Sis 2 Rz*- -.* *i V. rfhct.** Cv Kif ;*{< * M*r,tfst{t <u 'raj ttAt h'*n> ir\fn*?iH'Ty h) * M?-.>?..* J VsitelVv k. AnaruJ.KoheiU j. Mutpru X MonJlrMlWvr.li huttj',fid-* a f.r>::yM<jiw:*.*$?. sr> 4 M,..a a >- * Shac: -t.vr *K.'j}u<r-:.toj : pcdAine hkxsJicgdksrHlok*dinfncsluvr. * - }* i rn-f- V the t*:krf'fiy,'W . . ! ' Ji* IRI 1 \ VafW 1 It: a '. *v ** * V V ' N. l ) M-Ml AN ? V.isi: i VJ a h h ;-u :ih u 1 i a ^: Mi'UKMMtMOMIlsOI Mi. lV*V<K KWHS* 1 * ' 1 7l/ i vn'j 1 * <.V|U| .) ar.s-IMl ** * , 0 frij ,i * * 4 j t"H r .'> ; .- so- * i- t*> 7 r o -j j s *wa.lN V !< J ll-.'l.* 1 *.*' t>~ :! M I 0-0.11 J i t.n> TRANSMISSION OF AGENT OF POST-TRANSFUSION NONA, NON-B HEPATITIS BV (CRVOPRECIPITATE PREPARED FROM PLASMA OF SYMPTOAtLESS (CHRONIC CARRIER Sia.-Noo-A, non-B (NAKB) hepauiis after blood transfusion often differs clinically ftom that w hich deu-lups folluu mg infuston of (nttlucUKipliilic fatlut (factor"Vjll contt-juraic).1 Ctoss-cluUcngc studies in chimpanzees suggest that these clinical differences may reflect the exisience ofiwo different agents." To determine whether n agent itsponsihlc for ntany nr m<tst cases of blood transfusion associated HAND liepatins in the United States1 could also be lraitsmitt,ted by FVIII concentrate, which is purified from cryoprccipitate, we produced eryoprecipitate from plasma containing this agent. The plasma used (inoculum t) was that ofa symptomless patient who had acquired chronic NAKB hepatitis after transfusions and whose blood had tiansmincd NAKB hepatitis to than (via a broken capillary pipettelandtochimpanzees by experimental inoculation* This inoculum had a NAKB hepatitis infectivity litre ofabout IDO chimpanzee infenious doses per ml. 107 ml of inoculum 1 which had Ken stored continuously at -70X was thawed until the plasma was liquid at D'C. The cryoprccipitate wasrecovcied by cemiifugation at -591)0 g (model RC 2B, Sorvall, Korwalk. Connecticut) for 20 mm at 0CC. The yieldofcryoptccipiiait was I -1 g, Ir was iccor.M.-ijn J in lOmlofa solution of JO mol-l sodium citiate and J 35 mmoi l stvlium chloride at pH 7" 0. The volume of the reconstituted cryoprccipitate was about one-tenth that or the starting plasma and contained 10 mg pitucin.ntl 2 ir.! of reconstituted ciyopreupnats ii.oculatcd intravenously into a 1-1-month-olj female chimpanzee w.-ighing 8 kg. T!u Ji.ul Ken burr .* S bri-v*!- p .t-'-'-.y. !-.J had no ptiur exposure in Sources i-f hepatitis. t.*d not been previously inoculated wfitb any Koud, plasma, or plasma dertvat ives. Acute hepatitis, defined by lh. d.-itct.t--. iftvsistophilie Jig.r.i-ij'.ioi. of hepauvrstet and lyn.ptvo.yt.. . utictt. 1 was 1 g>..ltAf S' ilfir p A'ncswthicU u' ask.*.^:rS mith <rau*rntfi "..r.lt.W n(.ub>tV1li<(inTm:|l!t aw*-.r.' IV'ft s. ? Hu Io w t .') H-ManVa JR ..S2Htwnr'*tt<A*klMVl s.tiM .ii'firi Fl" l*rul\i5tor. lUlttnnlitiiAitiiKs tluils liijsciuiKtiu! (thltr f In' W ' dt. A, htut }t ht[q|il limn T )*' P* Ul:4IK'-<7 I) V*i11 k.iAis. R) Ak u i-(J turn-Vfc fw^aitits k'.toak .hdlknge ofsh:mron:rf) tl*. ih'tr Jr.l SwiiMRMlB ik* lW.I4C: 7**-V| 4 `Jdoti' 1 Vfl | H. tenu H! (Iitw.. mm A. r.o>.}> .mtri v a* tt*nm,ettkk grit' JiMtnxmrd ttiivtsrjk<trtil t'-f akst ssc pftu! tS (SA. .S4J. :n -4 l J i i I i i i lr [ >: ii i TEH 0532943 DUP050034213 16- Appendix III Chicago,. New York and Louisville Screening Data; Multiple Regression with Gasoline TEH 0532944 DUP050034214 -22- TABLE 4 (Continued) RESULTS FROM CPC SCREENING DATA Dependent Variable: Chicago, Black, Percent Over 30, Logistic 1 Sum of Squared Residual = 903.7910 Standard Error of the Regression -- 4.2947 Mean of Dependent Variable = -5.9293 "* Standard Deviation * 5.3271 Log of Likelihood Function * -155.0210 Number of Observations ** 55. 1 Sum of Residuals * -12.3123 Durbin-Watson Statistic (Adjusted for l.GAPS) = 0.6633 1 Right Hand Variable Estimated Coefficient Standard Error T-Statistic Constant SMSA Gas Lead Vinter Spring Summer Fall -2.06177 .200S26E-01 - .802201E-01 - .153203E-01 ,30173 - .20619 1.09922 323946E-01 1.08125 1.07987 1.07956 1.08049 -1.87566 6.19942 - .741917E-01 - .141872E-01 .27949 - ,19083 1 Dependent Variable: Chicago, White, Percent Over 30, Logistic Sum of Squared Residual = 236.7520 Standard Error of the Regression = 2,1981 Mean of Dependent Variable = -5-4757 5 Standard Deviation = 2.5221 St Log of Likelihood Function = -118.1630 Number of Observations = 55.0000 Sum of Residuals - -15.5591 Durbin-Watson Statistic (Adjusted for l.GAPS) - 0.9848 Right Hand Variable Estimated Coefficient Standard Error T-Statistic Constant SMSA Gas Lead I Vinter Spring Summer Fall -2,901420 .225653E-01 .964196E-01 .627356 * , 62619IE-01 - .965362E-01 .611533 .370419E-01 .566928 .563742 .562615 .562510 -4.74450 6.09163 .17007 .11128 - .11130 - .17078 TEH 0532345 & iCF Incorporated DUP050034215 21- TABLE A RESULTS FROM GDC SCREENING DATA Dependent Variable: New York, Black, Percent Over 30, Logistic Sum of Squared Residual = 1,565.410 Standard Error of the Regression * 6.994 Mean of Dependent Variable = -14.631 Standard Deviation = 9.06? Log of Likelihood Function = -130.098 Number of Observations = 40. Sum of Residuals = -25.960 Durbin-Watson Statistic (Adjusted for l.GAPS) * 0.704 Right Hand Variable Estimated Coefficient Standard Error T-Statistic Constant SMSA Gas Lead Early Tri-State Gas Lead Later Tri-State Gas Lead Winter Spring Summer Fall -1.49686 6,49250 .23089 - .92929 - .23515 - .14904 .31506 691338E-01 2.03605 2.33997 .20763 .55681 1.75445 1.75109 1.75327 1.75120 - .73518 2.77461 1.11200 -1.66896 - .13403 - .851138E-01 .17970 .394780E-01 Dependent Variable: New York, White, Percent Oyer 30, Logistic Sum of Squared Residual == Standard Error of the Regression 5= Mean of Dependent Variable Standard Deviation .35 Log of Likelihood Function Number of Observations .= Sum of Residuals x-'j *bin-Watson Statistic (Adjusted for l.GAPS) 96.7623 1.7389 -7 .5064 3.0394 -74.4251 40. -15.3735 1.3655 Right Hand Variable Estimated Coefficient Standard Error T-Statistic Constant SMSA Gas Lead Early Tri-State Gas Lead Later Tri-State Gas Lead Winter Spring Summer Fall -4.16557 5.09855 .59210 - .21366 - .155031E-01 - .14977 .544287E-01 .18870 1.03897 2,06430 .18678 .49413 .45508 .44233 .45301 .44248 -4.00933 2.46986 3.16996 - .43240 .340670E-01 - .338600E-01 - .12015 .42645 ICF Incorporated DUP050034216 TEH 0532946  _i ,;- iAMnnj!jiSitffrr-V r F5t -23- TABLE 4 (Continued) RESULTS FROM CDC SCREENING DATA Dependent Variable: Louisville, Black, Percent Over 30, Logistic Sum of Squared Residua) = 328.7350 Standard Error of the Regression ** 5.2340 Mean of Dependent Variable = -7.3389 I Standard Deviation -- 5.7024 Log of Likelihood Function * -51.6848 Number of Observations - 18. Sutn of Residuals -6.6824 Durbin-Watson Statistic (Adjusted for l.GAPS) ** 1.6956 Right Hand Variable Estimated Coefficient Standard Error T-Statistic Constant SMSA Gas Lead Vinter Spring Summer Fall 2.84155 33.79160 853627E-01 .18590 .161720E-01 .2874$ 1.46965 10,35040 1.32522 1.32784 1.33636 1.33014 -1.93349 3.26478 .644138E-01 .140039E-01 ,121016E-01 - .21613 *"*-**- ear r"' Sr m ' ' .4. :T u-4. v9* Dependent Variable: Louisville, White, Percent Over 30, Logistic Sum of Squared Residual = Standard Error of the Regression = Mean of Dependent Variable = Standard Deviation == Log of Likelihood Function = Number of Observations * Sum of Residuals - Durbin-Vatson Statistic (Adjusted for l.GAPS) = 97.94640 2.85696 -6.17638 3.04550 -40.78730 18. -5.62555 1.83940 Right Hand Variable Estimated Coefficient Standard Error T- Statistic Constent SMSA Gas Lead Winter Spring Summer Fall 3,07121 33.22030 .753012E-01 .16809 .335372E-01 .20986 * .936198 8.673070 .739037 .737968 .741039 .741126 -3.34482 3.37439 .10189 ,22778 - .452570E-01 - ,26316 TEH 0532947 ICF incorporated DUP050034217 OWM Energy and Resource Consultants, Inc. P.0. Drawer 0, Boulder, CO 80306 (303) 449-5515 THE USEFULNESS OF THE NHANESII DATA FOR DISCERNING THE RELATIONSHIP BETWEEN GASOUNE LEAD USAGE AND BLOOD LEAD LEVELS IN AMERICANS AND A REVIEW OF ICFS ANALYSIS USING THE NHANES H DATA Prepared for: Environmental Protection Agency Office of Policy Analysis March 1983 Contacts Dr. Daniel Violette Dr, Craig Miller Our Ref: MEMO-SCHAR (100) TEH 0532948 DUP050034218 --~------------ --- Energy and Resource Consultants, Inc. --------------- -------- MEMORANDUM The Usefulness of the NHANES II Data for Discerning The Relationship Between Lead in Gasoline and Blood Lead Levels in Americans and A Review of ICPS Analysis Using the NHANES II Data Date: March 4,1983 To: 3oel Seijjptz, Office of Policy Analysis, U.S.EPA From: Dr, Daniel Violette and Dr. Craig Miller This communication is in response to your request that we review the use of the NHANES II data in the December 1982 report entitled "The Relationship Between Gasoline Lead Usage and Blood Lead Levels in Americans: A Statistical Analysis of the NHANES II Data" prepared by IGF, Inc. As you know, our firm is currently working with the De partment of Biometrics at the University of Colorado on a project using the NHANES I and II data to examine potential relationships between hazardous air pollutants and health effects. In a previous project, Dr. Violette reviewed statistical studies relating air pollution to health effects for the National Commission on Air Quality. That latter work appeared in Environmental Policy: Air Quality, edited by G. S. Tolley and P. . Graves, and published by Ballinger in 1981, Dr. Miller's experience includes managing a statistical project examining the relationship between soft water and heart disease. With this experience, as well our participation in many nonhealth related statistical projects, we feel qualified to comment on the ICF study. The background documents provided to us depict a lively debate, although a good portion of the controversy appears to be fueled by misunderstandings of the statistical ap proaches used by the authors of the different papers. This is particularly apparent in the ICF comments (3) on the Ethyl Corporation Analysis (2) and the replies by Draper (5) and 1 TEH 0532949 DUP050034219 Energy and Resource Consultants, Inc. Bradley (6). We feel ithat it is important to point out that Draper's very critical com ments were confined to ICF's critique of Ethyl Corporation's analysis and were not directed at ICPs statistical analysis presented in either their October 1982 (4) or December 1982 (1) reports. Jh general, we feel that the ICF use of the NHANES II data is appropriate* the analysis is sound statistically and! the results provide useful and needed insights into the possible relationship between lead in gasoline and blood lead levels. As is true for any statistical study, improvements are possible, and we offer several suggestions in this regard. In the balance of this communication, we will address whether the NHANES II data can reason ably be used to study changes in blood lead levels over time, offer some general comments on the ICF analysis (1), and make a few comments on the Ethyl Corporation analysis (2) that we feel were not covered in previous critiques. We would enjoy pre paring more detailed comments on the statistical analyses performed to date, but it was not possible to do so and meet the requested delivery time for this communication. the first question concerns whether it is appropriate to use the NHANES II data to ex amine changes in blood lead levels over time, the NHANES II data was designed to be a cross-sectional--not a time-series study, the time trends that are available stem from the five-year duration of the sampling program. Over the five-year period, two sampling stations were used to obtain data at a number of locations, with each location being sampled once, the result is not a conventional set of time series data where there are observations on individuals at a single site in successive time periods, the data set contains information on individuals in different sites for the different time periods, this results in a "moving" sample over time, the general concern is that this data set does not constitute a representative random sample at each point in time and, therefore, could bias the statistical analysis. TEH 0532950 DUP050034220 Energy and Resource Consultants, Inc. After reviewing the data set and the comments by 'die different reviewers, it is our con clusion that it is perfectly appropriate to use die NHANES 11 data set as part of an effort to determine whether any trends in die data during the 1976 to 1980 time frame can be explained. However, use of an analysis based on this data to forecast trends for the nation must be done carefully and with foil recognition of possible sampling biases. We are drawing a distinction between an analysis of what is going on within the data set (which is perfectly appropriate) and using the data to forecast national trends, the validity of any forecasts based on the data must be judgmentally determined. If a strong relationship between lead in gasoline and blood lead levels is found, then the implications of this finding for the nation should be explored and the information provided to policy makers with the appropriate caveats. This second use of the data, i.e., forecasting, must be based on sound judgment, but it is surely an appropriate use. This is particularly true if this data set is one of the few available for examining the relationship between gaso line lead and blood lead levels. This being the case, the useful information available from this data should not be ignored. It is not appropriate to ignore die information from such a data set simply because it is not a representative random sample. Further, few data sets used in health studies are representative random samples in the strictest sense. The literature on health effects contains many examples where "natural experiments" were used. These analyses use data from several locations that have been subject to different environmental impacts. The choice of locations is not random, but stems from the availability of unique data. The data is analyzed stat istically with controls for as many potential biases as possible, and judgmental extrapolations to a larger population are made,. The use of the NHANES n data by 1CF is just as appropriate, if the potential biases are controlled. This conclusion seems to be shared by the different individuals who have been involved in the debate. Bradley (6) on pages 15 and 16 in reference to the NHANES II data states: 3 TEH 0532951 DUP050034221 rwry Energy and Resource Consultants, Inc. If it were possible to devise a planned, controlled experiment to directly examine the effect of lead in automobile emissions on human blood lead, we would not need a sample of the U.5. population but rather homogenous groups of individuals falling into specified personal characteristic groups subjected to controlled exposure to automobile emissions. Such an experiment seems impossible, but the Ethyl Corporation's examination of individual records with variable exposure because of gasoline lead use may be as close as one can come to the desired experiment. It is not necessary to have population estimates. It is only necessary to examine the rela tionship of blood lead to a measure of exposure after standardizing in so far as possible for other causes of variation. (Emphasis added.) The analysis by the Ethyl Corporation (2) states on page 21: ...there was a need to analyze the NHANES II data using techniques designed to reduce (or eliminate) the confounding of these variables, which Is inherent in the experimental design of the NHANES II study. To accomplish this objective, we designed a linear regression model that was used to eliminate the effects on blood lead (dependent variable) of per sonal characteristics (independent variables)--age, sex, race and family income. These statements by the parties that have participated in the debate indicate that use of the NHANES II data is appropriate as long as variables are incorporated to control for the potential biases that may occur due to the NHANES II sampling procedures. The objections that were raised by the Ethyl Corporation (2) and Bradley (6) were to the use of simple correllations between gasoline lead and blood lead levels that were used by Honk (7) and in other earlier analyses. They did not criticize the use of multivariate analysis of the NHANES II data, which can control for these potential biases. We fully Ufa.uv with this interpretation. The IGF Analysis This section will briefly discuss the December 1982 ICF analysis. The discussion of potential sources of biases in the models (ICF, p. 5) is quite complete, and the methods used to control for these potential biases (ICF, p.6) represent good statistical practice. ICF controls for the changes in the locations and mix of persons sampled over the period TEH 0532952 DUP050034222 Energy and Resource Consultants, Inc. of the NHANES D suivey by incorporating a variety of subpopulation explanatory vari ables (such as age, income, urban, etc.). This is standard practice and might be fully adequate for controlling for die potential biases. However, a more extensive set of Control variables could be used. In particular, 0-1 dummy variables for each site could be incorporated into the analysis. The incorporation of site dummy variables should not be restricted by the sample size, since on average, blood lead levels for 150 people were taken at each of the 64 locations. The incorporation of site dummy variables would capture the effects of variables specific to each location that may influence die blood lead levels of the individuals but were omitted from the equation. The dummies could be entered as additive terms, thereby allowing the intercept to vary for each site, or they could be entered as multiplicative interaction terms which would allow die coefficients on the independent variables (e.g., gasoline lead) to vary across sites. The use of site dummy variables should better control for potential biases due to die NHANES II sampling procedure.* Although the ICF analysis is sound, there were several points we found confusing that had to be cleared up in conversation with the authors. In examining die presentation of the ICF results in Table 5, we were uncertain whether they had used the 10,000 plus observa tions on individuals or had aggregated die data and used the average blood lead values for c each location as the dependent Variable, The source of the confusion was die 32 degrees of freedom associated with each regression run presented in Table 5 and the lad< of an explicit definition of the dependent variable used. With observations on thousands of individuals and only 15 independent variables in the equation, one would have expected a much larger number of degrees of freedom for die regression. The ICF authors explained that the use of the SURREGR program to adjust for the NHANES II stratified sampling * ICF has informed us that they are incorporating site dummy variables in their next analysis. 5 TEH 0532953 DUP050034223 -------- !" ----------- Energy and Resource Consultants, Inc.------------------------------- scheme accounted for the reduced degrees of freedom and that they had used the data on individuals. Incorporating an explanation of SURREGR in the report would allow for a better understanding of the analysis. We still are not quite sure if the 32 degrees of freedom listed in Table 5 constrains the number of variables that can be included in the regression equation or whether it is an independent constraint on SURREGR. A second suggestion is to better qualify the conclusions of the analysis, particularly as they appear in the Executive Summary* A reader becomes very wary when confronted with adamant and positive conclusions based on any statistical study. Qualifications are particularly necessary when presenting forecasts based on this data set. For example, it is assumed that all potential confounding factors in the sample have been removed, and this should be stated. In conclusion, the ICF analysis is sound, and the results--with appropriate qualifications-- should be considered as one piece of information useful for setting standards. Of course, more could have been done, but the analysis is typical in quality and approach of the statistical studies that appear in the published literature on health effects. Comparison to the Ethyl Corporation Analysis If the ICF analysis is sound and their finding of a stong positive relationship between the amount of lead in gasoline and measured blood lead levels reasonable, then what explains the difference in the results between the Ethyl Corporation's analysis and the ICF study? We believe the difference in the results can be at least partially explained by the different independent variables used by Ethyl to represent gasoline lead levels and the two-stage procedure they used. Both of these factors, in our view, bias the estimated relationship between lead in gasoline and blood lead levels in the direction of finding no relationship. TEH 0532954 DUP050034224 Energy and Resource Consultants, Inc. First, it is clear that the two-stage approach biases the estimated coefficients on gaso line lead levels towards zero. In the first step, Ethyl regresses site dummy variables, age level, race, sex and income (99 variables in all) against the mean blood lead levels. The residual variation (i.e., unexplained variation) from this first stage is then toe indepen dent variable in toe second stage where variables related to lead in gasoline are included. This two-stage approach assumes that any common covariation between toe second-stage gasoline lead variables and toe independent variables in toe first stage are attributable entirety to the first-stage variables in their explanation of toe dependent variable (he., blood lead levels). If lead in gasoline Is felt to be likely to influence toe blood lead levels, them omitted variable bias will be introduced into toe first stage regression. The severity of this bias will depend upon how closely the omitted gasoline lead variables are correlated with the variables included in toe first stage regression. The result will always be a downward bias in toe estimated coefficient in toe secondstage regression. The extent of toe bias will depend upon how closely toe gasoline lead variables are correlated with toe first stage variables. This determines how much of the variation in blood lead levels that could be due to variations in the lead in gasoline, is now falsely attributed to toe first stage independent variables. Ethyl shows in Figures 3, ft and 5 of their analysis that there is likely to be a strong Correlation between lead in gasoline and the age level since both have declined over time.2 Even if this were not the ers.a, regressing the 99 variables against toe blood lead levels would iikety explain a fair 2 There is another variable used by Ethyl in their first stage regression that might be Co.-related with toe lead in gasoline variable. Rather than using a set of sixty-three 0-1 dummy variables to control for individual observations taken at the 6ft different locativis, Ethyl assigned toe locations numbers from 1 through 6*1 and, apparently, used this variable in their regression (see page A-l of reference 2), It is not clear how these num bers were assigned to the locations. If they were assigned according to toe time the lo cation was sampled, with toe first location being 1 and toe last being 6ft, then an obvious time trend is introduced into the first stage regression that will be positively correlated with any variables increasing over time and negatively correlated with any variables de creasing over time. Even if npt ordered by time, it is not clear why one site should have a different weight (value) than another and an arbitrary weighting or trend is introduced as a variable in the regression. TEH 0532955 DUP050034225 --<-- --------------------- Energy and Resource Consultants, Inc.------- ----------------- portion of the variance, leaving a highly random residual to be explained in die second stage. As a result, Ethyl's statement on page A-3 of the appendix that "die correlation of the relationdip between gasoline lead and blood lead should be strengthened when the effects of age, sex, race, and income level are removed" is incorrect. Finally , the two independent variables used to proxy the amount of lead in gasoline in die second stage regression would appear to introduce sizable measurement errors. It is a mystery why Ethyl did not directly incorporate die data they had on die levels of lead in gasoline as an independent variable in the regression. Instead, Ethyl used population density and gasoline lead usage for die state divided by die area of the state as the inde pendent variables. Population density is used "since gasoline lead usage would be expected to be highly correlated with population, therefore population density would roughly be interpreted as an indicator of the gasoline lead usage per unit area and thus represents the local air-lead concentration exposure." Of course, there are many poten tial problems with this proxy measure. Sites that are less dense but have poor public transit such as those in the South and West may have higher lead exposures than areas with a higher population density due to a higher consumption of gasoline. Further, the amount of lead in gasoline has declined considerably over the 1976 to 1980 NHANES II sampling period while the population density at a location will have remained essentially constant. As a result, two locations with tee same population density would have very different exposures to lead in the atmosphere from gasoline consumption, if one were sampled earfy in tee NHANES II period and the other sampled late. Using tee population density as an independent variable cannot capture this change in lead content of gasoline over the sampling period and, therefore, is a poor proxy for exposure to lead from gaso line consumption. The other gasoline lead variable (state gasoline lead u$e divided by state area) is meant to provide a variable to represent lead fallout entering the food chain. One need only consider that much of the food consumed in any region is not grown in state and tee pro- TEH 0532956 DUP050034226 TUJsep**to*i Energy and Resource Consultants, Inc. portion of in-state food which is consumed locally will vary between states (eg., California versus Arizona). With the two-stage approach and die unusual proxy variables for lead in gasoline, it is not surprising to us that no significant relationship was found in die Ethyl analysis. Final Comments Any evaluation of the appropriateness of data and analyses for inclusion in a survey or summary document should take into account die body of research that has been accumu lated. The appropriate criteria for inclusion concerns whether the study under consider ation makes a positive contribution to die research, not whether die study is perfect in all respects. Obviously, it would have been desirable to have time series data for an unchanging set of locations but, with the incorporation of control variables in a multi variate regression analysis, the use of the NHANES II data Is appropriate. We feel that the ICF analysis does make a positive contribution to the literature and should be included in any compilation of research on the relationship between lead in gasoline and blood lead levels. Of course, a policy decision is never made on just one statistical study, but is based on the weight of the evidence which incorporates both laboratory studies and epidemiological studies. The ICF study is one piece of information that should be consi dered. As a final note, both Dr, Violette and Dr, Miller have reviewed many of the studies that were included in the document, "Air Quality Criteria for Particulate Matter and Sulfur Oxides--Volume IV, Health Effects," The appropriateness of the data set and die statis tical analysis used in the ICF study are of a quality comparable to the epidemiological studies included in chapter 14 of the above Criteria Document and, In fact, are better than many of the studies included in diet document. TEH 0532957 DUP050034227 Energy end Resource Consultants, Inc, REFERENCES 1) "The Relationship between Gasoline Lead Usage and Blood Lead Levels in Americans: A Statistical Analysis of the NHANES II Data," prepared for the Office of Policy and Resource Management, U. S. EPA by ICF, Inc.* December 1982. 2) Ethyl Corporation's Comments following EPA's Public Hearing on Proposed Rule for Lead in Gasoline, Washington, D. C., Septenriber 7-8, 1982 - Submitted to EPA's Docket No. A-81-36, October 8,1982. 3) "A Review of toe Ethyl Corporation's Comments on EPA's Proposed Rule for Lead in Gasoline," prepared for toe U. S. Environmental Protection Agency, Office of Policy and Resource Management, Office of Policy Analysis, by ICF Incorporated, November 1982, 4) "The Relationship between Gasoline Lead Emissions and Blood Poisoning to Americans," prepared for toe Office of Policy and Resource Management, Office of Policy Analysis, by ICF Incorporated, October 1982, 5) "Comments on the Documents: Ethyl Corporation's Comments Following EPA's Public Hearing on Proposed Rule for Lead to Gasoline and Review of Ethyl Corporation's Comments by ICF." by Norman Draper "for Ethyl Corporation, undated. 6) "A Discussion of Issues and Conclusions on Gasoline Lead Use and Human Blood Levels" by Ralph A. Bradley, for toe Ethyl Corporation, undated. (7) Statement by Dr, Vernon N. Houk, Acting Director, Center for Environmental Health, Center for Disease Control, before toe Subcommittee on Environment, Energy and Natural Resources, Committee on Government Operations, U.S. House of Representatives, Washington, D.C., April 14 1982. 10 TEH 0532958 DUP050034228 * Energy and Resource Consultants, Inc, r ef er enc es 1) "The Relationship between Gasoline Lead Usage and Blood Lead Levels in Americans: A Statistical Analysis of the NHANES II Data," prepared lor the Office of Policy and Resource Management, U. 5, EPA by ICF, Inc., December 1982. 2) Ethyl Corporation's Comments following EPA's Public Hearing on proposed Rule for Lead in Gasoline, Washington, D. C., September 7-8, 1982 - Submitted to EPA's Docket No. A-81-36, October 8, 1982* 3) "A Review of the Ethyl Corporation's Comments on EPA's Proposed Rule lor Lead in Gasaline," prepared for the U. 5. Environmental Protection Agency, Office of Policy and Resource Management, Office of Policy Analysis, by ICF Incorporated, November 1982. H) "The Relationship between Gasoline Lead Emissions and Blood Poisoning in Americans," prepared for the Office of Policy and Resource Management, Office of Policy Analysis, by ICF Incorporated, October 1982. 51 "Comments on the Documents: Ethyl Corporation's Comments Following EPA's Hf.arinr on Proposed Rule for Lead in Gasoline and Review of Ethyl Corporation's Comments by ICE," by Norman Drape- for Ethyl Corporation, undated. 6) "A Discussion of Issues and Conclusions on Gasoline Lead Use and Human Blood Levels" by Ralph A. Bradley, for the Ethyl Corporation, undated. 9 TEH 0532959 DUP050034229